How Performance Art Used AI to Locate BMW’s Hofmeister Kink

Unknown to most, the Hofmeister Kink has been an ever-present design element in almost every BMW model since its first appearance in 1962. Characterised as the slightly accented ‘C’ shape at the back of the rear passenger window, the Kink is celebrating its 60th anniversary in style - and with a little help from some seriously advanced tech that would have been unthinkable when it was first designed.

Developed through the combined efforts of Performance Art’s creative, strategy and technology teams, the agency has created an AI-driven platform that has found over 56,000 roads in the US that resemble the shape of the Hofmeister Kink, using machine learning to do so. At Hofmeisterkink.com, drivers can discover these roads, match them against the Kinks on their vehicle, or a vehicle of their choosing from BMW’s digital library, and even keep a shareable log of the Kink-shaped roads that they’ve driven on.

Speaking to LBB’s Ben Conway, Performance Art’s chief creative officer Ian Mackenzie discusses how the agency used the convolutional neural network YOLO v4 and MapBox API to create the interactive map that reveals Hofmeister Kinks on American roads and how the cross-functional team applied a “narrative approach to technology” to achieve it.

LBB> Where did the initial creative spark for this campaign come from? What was the brief from BMW like?

Ian> When we first started working with BMW in 2018, a few of us took part in some incredible brand onboarding, led by our clients at BMW. During that onboarding, we learned about the Hofmeister Kink. It shows you the value of good onboarding. By sharing literally every nook and cranny of the brand story with us, BMW got us invested and woven into the fabric of its brand. I personally had not known of the Hofmeister Kink before then. 

It stuck with us, and a few years later we wondered if there was a way to collaborate with AI to find that iconic Hofmeister Kink shape out in the world. That was the project’s inception. From there, it was looking for creative applications of that idea. As far as the brief goes, BMW is always looking for ways to deepen engagement with its customers – and together we built out the platform with that in mind. 

LBB> What is that client relationship like with BMW? How involved are they in the creative process - did their designers and creatives have much input on this project?

Ian> We’ve had the good fortune to work with BMW across North America in various capacities since 2018. This project was a collaboration between BMW’s American CRM team, a small group of key partners, and our agency team. This idea is at its best when it’s connecting across many technological, strategic and creative pillars – and required deep collaboration with our clients, led by the visionary Andreas Walter, GM customer experience and digitalisation at BMW of North America.

LBB> Where did the idea for the road-mapping tech come from? Was it a case of finding a concept that worked with the ‘kink’ design? 

Ian> We started with that hunch, that if you could train a visual AI to look for Hofmeister Kink-shapes in the wild, it would come back with interesting and maybe unexpected results. Our hypothesis was that we could literally find this iconic brand asset hiding in plain sight - we imagined it might find bridges or geological features that matched. So we briefed the creative team of Andrew Bernardi (CD) and Jordan Gabriel (ACD) on finding an executional way to bring the idea to life. One of the things they came back with was using the process to find road shapes that match – which we immediately fell for and began pursuing with vigour. From inception to launch, it took maybe a year. But if you think back to that original 2018 brand onboarding, it obviously took much longer. 

LBB> Who did you work with on the actual AI tech? Can you briefly explain how the mapping actually works and how you trained the AI?

Ian> We built this AI solution internally at Performance Art. We started by manually tracing dozens of Hofmeister Kink shapes from photos of BMW vehicles past to present. Then our team developed an algorithmically generated synthetic training set that produced hundreds of minor variations of each shape to give the learning model the highest chance of finding roads that match (because it turns out if you’re looking for a mathematically identical shape, you’re unlikely to find a perfect match). 

From there, we developed a machine learning model that ingested the synthetic training set and used it to find what it believed to be matching shapes within one square kilometre blocks of US map data. We ran localised tests and then, based on the outputs of the tests, tuned the model to improve its accuracy. And finally, we ‘hand-pruned’ out kinks that the model found to be a match, but the human eye did not. There’s a much longer answer in here, but we’re proud of the robustness of this solution. There are probably other, lesser ways to find a few Hofmeister Kink-shaped roads. Our approach and solution prioritised scalability, and was guided by our tech team to push at the limits of our industry’s capabilities with machine learning - versus just a vanity use of AI for the sake of a marketing halo.

LBB> How did you decide to display this as a map and to allow people to track whether they’ve driven these ‘kinks’ before? What does this add to the campaign in terms of engagement etc?

Ian> Throughout the process, we were always looking for ways to close the gap between the Hofmeister Kink and the driving experience. So, in addition to this amazing experience of actually finding the roads in the first place, how can we get people to consider driving them? Because, as it turns out, a Hofmeister Kink-shaped road is a great shaped road to experience a BMW’s road-hugging driving dynamics. So giving users a little nudge to log the ones they’ve already driven helps them connect to an experience they’ve already had - versus trying to get them to get in their car and drive one cold – which they’ve reported doing also.

LBB> Do you have any data you can share with us regarding how many people have tracked their driven locations? Or any other data you collected from this campaign?

Ian> We still tracking, tuning and driving folks to the site through paid media and CRM, but early results have been great. It’s a mix of people who are learning about the Hofmeister Kink for the first time, and those who already love this part of BMW’s design story. Some people are posting on social about the roads they’ve driven. Whether it’s just seeing the closest Hofmeister Kink-shaped road that matches your vehicle, tracking the ones you’ve driven, exploring other models, or browsing the map to see them all, there are so many ways to engage with the platform. We love them all, especially the ones that get people into their cars and drive on these roads. 

LBB> Have you worked with YOLO v4 or other AI/machine learning systems before? What is the key to the successful use of AI in a campaign? How do you stop it being just a gimmick?

Ian> We have worked with AI before. In general, we often start with something I’ve referred to as ‘a narrative approach to technology’ – in which we broadly imagine something to be possible without knowing that it actually is. For example: ‘From what I know about AI, I bet it could learn what a Hofmeister Kink shape is, and then find similar shapes in the wild.’ You don’t have to know much about AI to have that idea. 

Then, we bring in people who actually know how to make it happen. In this case, our amazing group director of product development/technology solutions, Arnaud Icard, developed the machine learning model and all the backend magic required to bring it to life. 

LBB> The use of MapBox API is an engaging aspect of the project - seeing the satellite image of the kinks adds that bit of reality to the map - what was it like working with this technology?

Ian> Key to the whole process was building a fully geo-enabled dataset. This allowed us to pull the data seamlessly across the MapBox platform and the additional GIS technologies we used to bring the experience to life. For the site, CRM and launch film, we sourced high-resolution aerial photography from Nearmap, providing stunning shots of the best kinks across the country. Because the data was geo-enabled, it was more easily able to flex across multiple platforms and uses.

LBB> How has this project inspired your future campaigns? And are there plans to expand this one?

Ian> Too soon to say, but we all have always dreamed of that roadmap. We’re in an exciting time where if you can dream it, and it’s a good idea, and you have a great team and smart clients, and you’re willing to go on the journey together, there’s just so much you can do – and we’re just right at the beginning.

LBB> What was the hardest challenge you faced on this project - and how did you overcome it?

Ian> Not sure it was a particular challenge per se but this project is a great example of a cross-functional team working together toward a singular objective. While the idea is clear and singular, there’s nothing simple about the execution. It requires the seamless integration of CRM systems, a large web build, customer data, multiple technology partners, business and strategic objectives, UX, front-end design, brand storytelling and well beyond. Getting it to work all together as well as it does is a wonderful achievement for the team.